The realistic learning environment
In previous chapters of this book, you learned about the general concepts and methodology of ML. While the information you saw gives enough technical background for approaching any learning problem, it does not suffice to tackle the security issues of ML models. We will therefore start with a discussion about the realistic setup of AML, which focuses on model security. In the interest of brevity, we will start by discussing an arms race problem in the context of cybersecurity.
Arms race problem
In the area of cybersecurity, a scenario reminiscent of an arms race frequently unfolds. This involves strategic competition among various entities, each vying for dominance over specific assets to fulfill their unique objectives. Classic examples of this dynamic include the ongoing struggle between malware developers and antivirus software engineers, as well as the contest between spammers and anti-spam technologies. This paradigm is also reflected in...
